Hacker News new | past | comments | ask | show | jobs | submit login
Towards an optical FPGA – Programmable silicon photonic circuits [pdf] (arxiv.org)
121 points by godelmachine on July 9, 2018 | hide | past | web | favorite | 38 comments


I'm the author of this paper [15] cited in OP's publication. We are the author of the very first implementation of a fully programmable optical linear circuit in silicon. This is the basic building block to have an optical processor (in this specific architecture).

Ask me anything.

[15] - http://www.photonics.intec.ugent.be/download/pub_3834.pdf

Fantastic work. Some off-hand questions, total noob in your field but interested in it and reading a lot:

- how fast is it (4KHz mentioned seems a bit slow, or is that a function of size?)?

- how far away from practical applications do you think you are?

- is the optical circuitry limited by the interaction with the electronic parts?

- the paper is already two years old, have there been interesting developments since then (in your lab, or elsewhere)?

The concept of speed that we use in a digital computer does not automatically applies to our optical circuit once we don't work using cycles of a clock. It's not a digital computer, but an operator that realizes a linear operation on an input. The speed of this operation is only bounded by the speed of the light propagating though the circuit.

We have optical chips in operation everywhere for decades, that's not something new. An programmable optical chip has been demonstrated by different research groups, and, for some limited applications, I do believe that we will have it in the marked in 7 to 10 years.

We are also working on different topologies besides the one mentioned in the paper.

Thank you!

Could you explain, or maybe point to a good textbook with a chapter on the subject, why the coupling efficiency to the second waveguide on the directional coupler peaks at a certain intermediate waveguide length? I have a PhD, just not in photonics.

I'll try an ELI5:

When the light propagates in a waveguide as the one used in my circuit, the E and H components of the EM wave are not fully confined to the waveguide, but part of it stays outside the waveguide. If you put two waveguides close to each other and makes the light travel to the first waveguide, part of the EM field of the light will also see the second waveguide.it makes part of the ligh couple to the second waveguide. As the wave travels, more and more light couples to the second waveguide. If you engineer it well, at some point 50% of the light will be confined in each waveguide. At this point you separate both WG and you have a 50:50 coupler.

Yes I know that the mode isn't perfectly confined in the waveguide, but Figure 4(b) of the article shows the coupling efficiency to the second waveguide going down as the length of the intermediate waveguide gets longer past some critical value. That's the part that gets me.

Pretty decent introduction if you have some physics background with some basic optical/E&M theory (PhD won’t help otherwise) is Silicon Photonics Design by Chrostowski.

How is the optical switching power needs being addressed?

Last I heard, you needed a high-powered laser on a non-linear material to turn transmission of a separate light signal on/off. Is that how things still work?

The chip described in this paper is actually less like a normal FPGA in that the chip isn't full of logic gates that can be reconfigured. Instead, it's full of optical elements (like modulators, multiplexers, beam splitters, and phase shifters) that can be connected in a certain pattern by erasing certain elements in the design.

Their hope is that by building a chip with lots of generally useful elements, many different industries will find it useful because they can just program the chip to utilize the components they need. (Fig. 6) Instead of computing with this chip, industry and researchers will be more interested in making things like optical transceivers. Having one common, reconfigurable chip that everyone uses would mean it could be mass produced at a lower cost, making research and production of silicon photonic systems more accessible.

At least that's what I've gathered from the paper - I'm definitely not an expert here. In theory I guess you could put optically switched gates in there too to make something more like an electrical FPGA - I'm sure that's long-term goal for the authors.

One main point to take for the analyses is that photons are bosons, so they don't interact between each other, which makes optical switches a pretty hard thing to do.

What we do is using interference to make a switch. More specifically, we use an apparatus called Mach-Zehnder interferometer (combined with a phase shifter) to manipulate the light.

Could you foresee fully photonic hardware at some point so not just silicon but motherboards, graphics cards and even external interconnects?

Is it possible to build RAM and SSDs in optical circuits yet? I ask because this has very interesting military uses that could push this forward if you could have devices impervious to electro-magnetic pulses.

I would think some of the first fully optical photonic devices would be network routers/switches. Stock traders have a lot to gain by shaving off microseconds. Possibly a fully photonic router could have a huge performance increase in the volume of packets per second and throughput as well?

I don't see photonics doing the data processing anytime soon. For that our current electronics works just fine.

What we do with photonics is to improve areas where the electronics is not efficient.

Case and point: interconnections. If you want to move data from point A to point B (being A and B either two different chips in your board our two datacenters), we can do it using electric signals, it works just fine, but at some cost. Electric signals dissipate power when they travel through a conductor, no matter how good the conductor is.

If instead we use optical signals instead of electrical, we have advantage in a number of points, specially power efficiency.

I want to build a small, personal photonics lab. What do I need?

Just out of curiosity, what would you aim to achieve/build with a lab like that?

I have some theoretical understanding of photonics but what sucks is that with programming, you just fork shit and start hacking. With physics, just starting hacking is much harder.

For starters I would try to make a hologram. Once you can do that you can expand into photonics, it means that at least your basic understanding of setup and things like interference and stability are taken care of.

That’s a good insight. Is there like some progression of this?

Hehe, it starts with a sand table...

Or maybe with reinforcing your floor joists :)

So maybe carpentry would be a good start?

Lol, I’m not sure understand. Is my comment of the base?

No, not at all, it's just that a proper optical setup is totally insulated from the surroundings so that a truck moving by 100 yards from your house doesn't wreck your carefully calibrated set-up.

Which usually translates into a lot of mass with as much vibration isolation as you can get.

And that in turn translates into large bins filled with sand (if you want to do this on the cheap).

And that in turn means you will most likely have to re-inforce your floor if you are in an ordinary building.

The best places to do work like this are in the middle of nowhere and away from any major roads. Even footsteps can be a problem for long running experiments where coherence is required.

It is a ton of fun though, and you will learn a lot.

Should I get learn zemax? Is that the right direction?

That's definitely going to teach you lots but I always loved to tinker and learn at the same time.

How about this?:


OCW should be archived.

I need help with the tinkering platform. I’m aware of the ocw class however it’s idk, I wanna build shit.

Warning: gateway drug, do not open package.


Because of course it will start with something simple... and then you start to wonder about larger and larger holograms.

More powerful lasers...

Bigger bases... (see sand table above)

Have fun!

How far are we from just using photon to run our computer instead of primarily relying to electron/electricity?

Is there anything that prevent a turing complete system that uses light/optic/photon to run instead of primarily using electron/electricity?

Not even close.

Photonics chips and programable photonics are not a way to substitute conventional electronics, but a replacement/complement for some areas of the field where electronics is not efficient.

I'm not very familiar with the subject, but I've been hearing about academic work on optical integrated circuits for a few years now. Is this work being commercialized yet? Are there photonic ICs in use in industry?


Photonics ICs are all around. You can find it in many different applications, but mostly in datacom and telecom.

A very simple example: anywhere you have a optical fiber you also need a photonic IC to, at least, convert the light into electrical signal to interface your electronics (and vice-versa).

What we are doing with photonics is adding more functionality in the optical part (filtering signal, multiplexing, modulation, etc) once, in mostly cases, it's more efficient doing that in the optical domain instead of using electronics.

How large is your proof of concept implementation? Do you think the architecture could be miniaturised enough to make a whole processor?

The full optical device is no bigger than 1mm x 0.5mm. It's very tiny in total size.

The point here is that our current implementation has a limited IO capability (our linear operator has 4 inputs and 4 outputs), but increasing the number of IOs leads to a linear increase in the size of the device.

What is the perceived potential in silicon photonic relative to traditional silicon in terms of:

* energy efficiency

* circuit density

* switching speed

* manufacturing cost

You're changing the world for the better.

I used to work in the industry and caught this article a few days ago during my usual arXiv trawling. Prof Reed's group has put out a few papers on erasable ion-implanted waveguides over the past couple of years from different angles, most notably erasable optical test points in circuits. The "FPGA" part is a bit of marketing because the waveguide erasure cannot be undone, but I do believe the last bit where the circuit topology is permanently altered between different optical communication protocols has some merit. This would enable manufacturers to be more nimble about which chips they can deliver based on demand, and might offer a yield recovery mechanism in a similar manner to which fuses are used on logic chips to disable non-functioning parts of the chip during the chip "Sort" step.

Overall, good paper.

There are still anti-fuse FPGAs in manufacture and in use today (one-time programmable burn-in devices)

My question: could programmable waveguides be done with MEMS technology? (mechanical features etched in silicon)

With "erasable waveguides" this seems more an "eASIC" than an FPGA, no?

There was a huge amount of interest, and an equally massive grant(billion) given to the Rochester,NY area for building up photonics research. Haven't heard much of it the past year though. Sounds like the natural progression from electrical circuits, particularly in the low thermal output.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact